PEW: Prediction-Based Early Dark Cores Wake-up Using Online Ridge Regression for Many-Core Systems

Future many-core systems need to address the dark silicon problem, where some cores would be turned off control chip’s thermal and power density, which effectively limits performance gain from having a large number of processing cores. Task migration technique has been previously proposed improve system by moving tasks between active As task imposes overhead due wake-up latency cores, this paper proposes prediction-based early (PEW) reduce cores’ during migration. A window-based online ridge regression (RR) is used as prediction model. The model uses past window’s thermal, power, core status (i.e., or dark) predict future temperatures at run-time. If predicted in next period, PEW puts state with low latency. Thus, reduces time for start executing tasks. comparison results show that our completion up 7.9% 4.1% compared non-early (NoEW) fixed threshold (FEW), respectively. It also shows increases MIPS/Watt 5.5% 2.3% over NoEW FEW, These improves system’s overall terms reducing increasing executed instructions per Watt.